OpenAI unveils ‘Sora’, a revolutionary AI that transforms text prompts into videos


    16th February 2024 – (New York) OpenAI has introduced a new tool capable of generating videos from simple text prompts. Christened ‘Sora’, meaning ‘sky’ in Japanese, this advanced model is pushing the boundaries of artificial intelligence, offering users the ability to craft footage up to one minute in length, which can adhere closely to their specified content and stylistic guidelines.

    The unveiling of Sora, detailed in a company blog post, marks a significant leap forward in the AI domain. The tool is not confined to generating novel content; it can also take a still image or a snippet of existing video and weave new visuals that extend the narrative seamlessly.

    Amidst the buzz, OpenAI has taken a cautious approach, granting access to Sora selectively, focusing on researchers and video creators who will rigorously assess the product. These ‘red teams’ are tasked with identifying any potential misuse, especially that which might contravene OpenAI’s established content policies. These guidelines are stringent, forbidding the creation of material that involves extreme violence, explicit content, and the unauthorized use of intellectual property.

    The strategic release of Sora to a limited user group comes with a commitment to responsible AI usage. The company’s chief executive, Sam Altman, took to social media platforms, engaging with users by sharing sample video clips generated by Sora, each bearing an AI watermark, further highlighting the tool’s capabilities.

    OpenAI’s portfolio of generative tools has seen rapid expansion, with the still image generator Dall-E launched in 2021 and the AI chatbot ChatGPT released in November 2022, which amassed a user base of over 100 million in record time. The introduction of Sora is set to complement this suite of AI-driven creativity tools, differentiating itself with the ability to produce more extended and contextually coherent video sequences than its contemporaries.

    While other AI entities have ventured into the realm of video generation, their offerings have often been limited to brief clips lacking substantive relevance to user prompts. Tech behemoths like Google and Meta are reportedly in the trenches developing their own generative video tools, yet no public releases have been forthcoming.

    In the backdrop of these advancements, OpenAI remains tight-lipped about the volume of footage and the sources utilised to train Sora. The New York Times received a statement indicating that the training corpus included a mix of publicly accessible videos and content licensed from copyright holders. OpenAI has previously faced legal scrutiny over its training methods for generative AI tools, which digest and replicate vast quantities of web-scraped data.